Scientists that use the workflow paradigm for the enactment of scientific experiments need support for trial-and-error modeling, as well as flexibility mechanisms that enable the ad hoc repetition of workflow logic for the convergence of results or error handling. Towards this goal, in this paper we introduce the facilities to repeat partially or completely running choreographies on demand. Choreographies are interesting for the scientific workflow community because so-called multi-scale/field (multi-*) experiments can be modeled and enacted as choreographies of scientific workflows. A prerequisite for choreography repetition is the rewinding of the involved participant instances to a previous state. For this purpose, we define a formal model representing choreography models and their instances as well as a concept to repeat choreography logic. Furthermore, we provide an algorithm for determining the rewinding points in each involved participant instance.